Page 14 - 《应用声学》2023年第2期

第 42 卷 第 2 期 Vol. 42, No. 2
2023 年 3 月 Journal of Applied Acoustics March, 2023

⋄ 研究报告 ⋄

一种深度学习的立体阵波达方向估计方法 ∗

刘兢本 1,2 郭良浩 1† 董 阁 1 闫 超 1

(1 中国科学院声学研究所 声场声信息国家重点实验室 北京 100190)
(2 中国科学院大学 北京 100049)
摘要：针对常规波束形成主瓣宽且目标分辨能力低的问题，提出一种基于深度卷积神经网络的波达方向估计
方法。算法使用常规波束形成计算二维空间功率谱，将预处理后的空间功率谱图输入深度卷积神经网络。该
文利用神经网络学习解卷积映射关系，输出主瓣宽度更窄的空间功率谱图，从而实现高分辨率二维波达方向
估计。该算法对阵列结构没有限制，适用于立体阵。仿真结果表明该文方法在不同目标个数、快拍数及信噪比
参数下均能准确估计目标方向。该文方法目标分辨能力优于常规波束形成方法。在低快拍情况下，目标方向
估计误差低于自适应波束形成方法。
关键词：二维方向估计；深度学习；神经网络；高分辨率
中图法分类号: TB566 文献标识码: A 文章编号: 1000-310X(2023)02-0202-15
DOI: 10.11684/j.issn.1000-310X.2023.02.002

A direction of arrival estimation method for stereo array

based on deep learning

LIU Jingben 1,2 GUO Lianghao 1 DONG Ge 1 YAN Chao 1

(1 State Key Laboratory of Acoustics, Institute of Acoustics, Chinese Academy of Sciences, Beijing 100190, China)
(2 University of Chinese Academy of Sciences, Beijing 100049, China)

Abstract: Aiming at the problem of wide main lobe and low target resolution in conventional beamforming,
a direction of arrival estimation method based on deep convolution neural network is proposed. The algorithm
calculates the two-dimensional spatial power spectrum using conventional beamforming, and feeds the prepro-
cessed spatial power spectrum into a deep convolutional neural network. In this paper, a neural network is
used to learn the deconvolution mapping relationship and output a spatial power spectrum with a narrower
main lobe width, thereby realizing high-resolution two-dimensional direction of arrival estimation. The algo-
rithm has no restrictions on the array structure and is suitable for stereo arrays. Simulation results show that
this method can accurately estimate the target direction under different target number, snapshot number and
signal-to-noise ratio parameters. The target resolution capability of the proposed method is better than that
of conventional beamforming. In the case of low snapshot, the target direction estimation error is lower than
that of the adaptive beamforming method.
Keywords: Two-dimensional direction estimation; Deep learning; Neural network; High resolution

2021-12-27 收稿; 2022-05-18 定稿
国家自然科学基金项目 (11874061), 中国科学院声学研究所自主部署 “目标导向” 类项目 (MBDX202105)
∗
作者简介: 刘兢本 (1995– ), 男, 湖北十堰人, 博士研究生, 研究方向: 信号与信息处理。
† 通信作者 E-mail: glh2002@mail.ioa.ac.cn